Current Projects

Show All

Novel algorithms for extraction of network models from multidimensional genome-scale data

We are interested in development of novel computational algorithms for automated extraction of complete network models from multidimensional genome-scale data, such as transcriptomics, copy number data, methylomics, metabolomics and proteomics. We have developed a biased random walk method, NetWalk, for effective scoring and querying of molecular networks from genome-scale data. NetWalk has been used in multiple projects in our own lab and in collaboration with others. We are currently working on extending NetWalk to analyses of multi-scale abstract networks for integration of multiple data types to build semi-abstract and interactive network models.

Novel algorithms for extraction of network models from multidimensional genome-scale data.

Figure 1. Comparison of A) list-based methods of network construction and B) NetWalk. In list-based network construction, interacting genes (open nodes) are added to the network of seed nodes (red) to connect them together. This will generate a single or a number of networks of interest. Distribution of data values of interactor nodes are random. In contrast, NetWalk transforms gene-centric data to interaction-centric data, which can be used for standard statistical analyses (e.g. heatmap analyses) or for dynamic network construction. Data values of nodes constructed through EF values are coherent with input values. 

Developing NetWalker: a software platform for network analysis in functional genomics

We are developing a software suite, NetWalker, as a freely available tool for the research community for conducting network-based analyses of genomic data. NetWalker features standard methods for basic data handling and processing, clustering and heatmap analyses; but more importantly, it features advanced methods for network-based analyses of genomic data in a highly interactive visual environment.

Developing NetWalker: a software platform for network analysis in functional genomics.

Analyses of molecular networks of multicellular communities

Cell behavior in vivo is shaped by the multicellular environment it resides in. To facilitate delineation of the molecular structure of the multicellular environment, we have extended our NetWalk algorithm to analyses of genomic data from multicellular systems. Using this method, we can analyze the molecular networks of multicellular communities that shape the collective phenotypic behavior of the whole multicellular system. We are currently engaged in a collaboration with Nancy Ratner's lab in the analyses of molecular networks in the highly heterogeneous multicellular microenvironment of Neurofibroma tumors.

Analyses of molecular networks of multicellular communities.

Figure 3. Community molecular network (CMN) analysis of tumor and stromal gene expression in drug resistance. A) Highest scoring interactions from the NetWalk-based CMN analysis of tumor and stromal gene expression. Nodes are colored according to gene expressions in tumor or stromal cells. Some subnetworks of intercellular interactions are highlighted in boxes. Gray interactions are protein-protein, and orange interactions are metabolic. B) A blow-up of the network 1 in A. C) A blow-up of the network 7 in A. D) Metabolic re-actions performed by enzymes in C and in subnetwork 9 in the tumor and stromal cells. Note a potential metabolic symbiosis between tumor and stromal cells in drug resistance involving division of glycolysis (e.g. glucose uptake, lactate production) and gluconeogenesis (lactate oxidation, pyruvate synthesis) between tumor and stromal cells, respectively.

Computational strategies for identification of onco-requisite networks from large cancer patient datasets

Targeted therapy of cancers aims to exploit specific vulnerabilities of tumor cells. A given oncogenic context of a tumor cell (i.e. collection of oncogenic mutations) will create specific vulnerabilities that could be targeted therapeutically. We call these oncogenic context-specific vulnerabilities as onco-requisite networks, as these networks promote survival of tumors with the given oncogenic context. For example, it was found that TBK1 signaling is specifically required for the survival of KRAS mutant cancers (i.e. KRAS mutation is the oncogenic context), making TBK1 signaling a KRAS-requisite gene. Fortunately, significant amount of genome-scale data is available on many cancer types from thousands of primary tumor tissues, which may enable computational identification of such onco-requisite networks. Our goal is to develop efficient computational methodologies to query onco-requisite networks for different complex oncogenic contexts from large multi-dimensional genome-scale data.



Molecular architecture of ERBB2-driven tumorigenic platform in breast cancer

Using combined computational and experimental approaches, we are investigating the molecular architecture of the ERBB2-driven tumorigenic platform in breast cancer. Although EGFR/ERBB2 signaling is well-characterized, the molecular networks contributing to ERBB2-driven tumorigenesis are not known. We make extensive use of our computational network analysis tools to query large genome-scale clinical and cell line datasets, and experimental validation analyses on a panel of breast cancer cell lines to model ERBB2-driven tumorigenesis.